A three-dimensional structure workpiece weld positioning device based on DLP three-dimensional vision

Abstract

The present invention discloses a device for positioning welds of a three-dimensional structure workpiece based on DLP three-dimensional vision. The present invention realizes the positioning of welds of a three-dimensional structure workpiece through the cooperation of a DLP structured light camera and a welding robot. First, a DLP camera is mounted on the end of the robot to shoot the workpiece in a multi-perspective manner, and a multi-perspective point cloud registration is performed in combination with the joint posture information of the robot body and the hand-eye calibration coefficient to generate a complete point cloud model of the workpiece. Then, based on the complete point cloud model, a weld extraction algorithm is designed to obtain effective positioning of multiple welds on the three-dimensional structure workpiece. Finally, the welding gun posture corresponding to each weld is solved to plan the welding trajectory of the robot body. The present invention does not require a CAD model of the welding workpiece and is particularly suitable for the positioning of welds of medium and thick plate three-dimensional structural parts. It has the characteristics of being automatic, efficient, stable, highly accurate in recognition, and adaptable, and meets the flexible and intelligent operation requirements of welding robots.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot welding, and in particular relates to a device for positioning weld seams of a three-dimensional workpiece based on DLP three-dimensional vision. Background Art

[0002] Vision-based welding technology has greatly improved the automation level of welding robots. Based on the steps of automated welding operations, it can be divided into weld seam positioning and automatic weld seam tracking. While current research on automatic weld seam tracking technology is extensive and in-depth, research on weld seam positioning is limited to situations where a single weld seam is simply distributed. Current vision-based weld seam positioning technology struggles to cope with diverse and complex workpiece welding tasks, such as those involving complex three-dimensional workpiece structures, an uncertain number of weld seams, and varying weld seam spatial distributions. Summary of the Invention

[0003] In order to overcome the shortcomings and deficiencies of the prior art, the purpose of the present invention is to provide a three-dimensional structure workpiece weld positioning device based on DLP three-dimensional vision.

[0004] The present invention is implemented as follows: a device for locating weld seams of three-dimensional workpieces based on DLP three-dimensional vision, which realizes weld seam positioning of three-dimensional workpieces by cooperating with a DLP structured light camera and a welding robot. The welding robot includes a teaching pendant, a controller, and a body. The device includes:

[0005] The workpiece point cloud model generation unit is used to capture the workpiece from multiple perspectives using a DLP structured light camera mounted on the welding end of the robot body. This unit then performs multi-perspective point cloud registration based on the robot body's joint pose information and hand-eye calibration coefficients to generate a complete point cloud model of the workpiece.

[0006] A weld positioning unit, configured to obtain effective positioning of multiple welds on a model-free medium-thick plate three-dimensional structural component based on the complete point cloud model;

[0007] The welding trajectory planning unit is used to solve the welding gun posture corresponding to each weld and plan the welding trajectory of the robot body.

[0008] Preferably, the workpiece point cloud model generating unit includes:

[0009] A hand-eye pose matrix acquisition module is used to calibrate the poses of the DLP structured light camera and the robot, and obtain the hand-eye pose matrix of the DLP structured light camera relative to the welding gun at the welding end of the robot body;

[0010] The robot posture reading module is used to read the posture of the welding gun at the welding end of the robot body relative to the robot base coordinate system from the robot controller in different postures of the robot;

[0011] Point cloud coordinate transformation module, used to transform the workpiece point cloud captured by the DLP structured light camera at the current viewing angle into the robot base coordinate system;

[0012] The coarse registration module is used to control the robot to drive the DLP structured light camera installed at the welding end of the body to move to multiple designated positions to collect point cloud data. The robot posture reading module and point cloud coordinate transformation module are repeated for the collected multi-view point clouds to transform all the workpiece point cloud data captured from each view into the robot base coordinate system to complete the coarse registration of the workpiece model.

[0013] The precise registration module samples the obtained coarse registration point cloud model and uses the improved ICP registration algorithm to complete the precise alignment of the point clouds shot under multiple perspectives, thereby obtaining a complete point cloud model of the workpiece.

[0014] Preferably, the weld positioning unit comprises:

[0015] Point cloud data preprocessing module, used for performing straight-through filtering, noise filtering, and work platform point cloud removal on point cloud data;

[0016] Feature point set recognition module, used to filter out high curvature feature point sets for subsequent weld identification from the acquired complete point cloud model;

[0017] The feature point concavity and convexity analysis module is used to perform geometric morphology analysis on the neighborhood of each high-curvature feature point acquired, complete concavity and convexity judgment, and screen out the feature point set with concave attributes;

[0018] The feature line extraction module is used to extract multiple feature lines from the acquired concave attribute feature point set using the Ransac algorithm;

[0019] The feature line segmentation module is used to segment each feature line using the Euclidean clustering algorithm to obtain multiple point clusters with independent spatial positions;

[0020] The weld positioning module is used to cluster a certain point, project all points inside the cluster onto the characteristic line according to the characteristic line where the cluster is located, and obtain the two endpoints of the weld through a sorting algorithm.

[0021] Preferably, the point cloud data preprocessing module is specifically used to:

[0022] Through filtering: retain the point cloud within the limited range of each dimension in the original point cloud data, and eliminate redundant data such as background point sets;

[0023] Noise filtering: Use statistical filtering algorithms to filter point cloud data to resist interference from factors such as environmental noise, low-key system noise, and edge discrete points;

[0024] Work platform point cloud elimination: Use the Ransac algorithm to fit and eliminate the work platform point cloud where the workpiece is located, retaining only the point cloud model of the workpiece to be welded.

[0025] Preferably, the feature point set recognition module is specifically used to:

[0026] Covariance analysis: For any point p in the point cloud i , search for its support radius r support All neighboring points Q i ={q i1 ,q i2 ,…,q in}, and p i Set as the center and calculate the covariance matrix of the local point cloud:

[0027]

[0028] where d ij is the neighborhood point q ij To the center point p i The Euclidean distance of

[0029] Eigenvalue solution: Use SVD operation to decompose the covariance matrix M and find the corresponding eigenvalues ​​λ1, λ2, and λ3, where λ1>λ2>λ3;

[0030] Feature point setting: Calculate the value of λ3 / λ2 and compare it with the set feature threshold. If it is greater than the threshold, it is set as a feature point;

[0031] Feature point set screening: Repeat covariance analysis and eigenvalue solution to complete the screening of the workpiece feature point set.

[0032] Preferably, the feature point concavity and convexity analysis module is specifically used to:

[0033] Directed normal calculation: For a certain high-curvature feature point, we first perform principal component analysis on the point set within its R radius neighborhood, take the eigenvector corresponding to the minimum eigenvalue as the normal of the feature point, and redirect the normal direction to point to the camera viewpoint, thereby eliminating the bidirectional ambiguity of the normal.

[0034] Feature point centroid analysis: Calculate the centroid of the neighborhood point set of the feature point, and define the vector from the feature point to the centroid as the concave-convex direction vector d of the point;

[0035] Convexity determination of feature points: Calculate the angle between the feature point's orientation normal and the concave-convex direction vector d. When the angle between normal and d is greater than 90°, the feature point has a convex attribute. Conversely, if the angle between normal and d is less than 90°, the feature point has a concave attribute.

[0036] Traversal and repeated analysis: For other high curvature point sets, the process of oriented normal calculation, feature point centroid analysis, and feature point concavity and convexity determination is repeated in sequence to complete the concavity and convexity attribute determination of all feature points.

[0037] Preferably, the feature line extraction module is specifically used to:

[0038] Recording of feature line interior points: Randomly select two feature points from the extracted concave attribute feature point set for straight line fitting, and calculate the distance from other feature points to the fitted straight line. If the distance is less than the set threshold, it is recorded as the interior point of the feature line;

[0039] Fitting line analysis: cyclically repeat the recording of the inner points of the characteristic line, and record the inner points and their number of each fitting line;

[0040] Fitting and elimination: Select the result with the largest number of inliers during the iteration process to complete the fitting of this round of feature lines, and eliminate the inliers of the feature lines from the original feature point set;

[0041] Traversal fitting: The process of recording the points inside the feature line, analyzing the fitted line, and fitting and eliminating points is repeated continuously until the number of points inside the fitted feature line fails to meet the set minimum threshold, and the fitting of all feature lines is completed.

[0042] Preferably, the feature line segment segmentation module is specifically used to:

[0043] Inclusion into clustering: For a set of points within a straight line, randomly select a seed point and search the neighborhood of the seed point within a radius of R. If there are other points in the neighborhood, include them and the seed point into the cluster Q;

[0044] Traversal clustering: randomly select new seed points in cluster Q and repeat the process of clustering until the number of points in cluster Q no longer increases, then the clustering ends;

[0045] Multi-cluster inclusion: New seed points are randomly selected from the remaining point cloud, and the process of inclusion, traversal, and multi-cluster inclusion is repeated to divide the original point cloud into multiple spatially independent clusters.

[0046] Preferably, the weld positioning module is specifically used for:

[0047] Set projection: For each cluster point cloud cluster obtained by segmentation, the internal point set of the point cloud cluster is projected onto the line according to the fitting parameters of the line to which it belongs;

[0048] Solve the current projection endpoint: Sort the projection point set on the projection line and find the starting and ending projection points of the point cloud cluster in the direction of the line. These two projection endpoints are the endpoints of the weld.

[0049] Solve all projection endpoints: Repeat the process of collective projection and current projection endpoint solution to complete the confirmation of all cluster point cloud cluster projection endpoints, and then all weld endpoints can be solved.

[0050] Preferably, the welding trajectory planning unit includes:

[0051] The discrete point sampling module is used to complete the sampling of discrete points of a certain extracted weld line segment starting from the initial point of the line segment with a fixed step size;

[0052] Discrete coordinate system construction module, used to establish a discrete coordinate system relative to the robot base coordinate system at each discrete point. The coordinate system consists of a normal vector, a direction vector, and a proximity vector;

[0053] The welding posture solving module is used to establish a welding gun coordinate system at the welding gun tip position, and calculate the welding posture of the welding gun at each discrete point of the weld according to the corresponding relationship between the weld posture and the welding gun posture;

[0054] The motion trajectory acquisition module is used to perform least square fitting on the welding gun trajectory to obtain the motion trajectory of the welding gun during welding of each weld.

[0055] Compared with the shortcomings and deficiencies of the prior art, the present invention has the following beneficial effects: the present invention does not require a CAD model of the welding workpiece, is particularly suitable for the weld trajectory of medium and thick plate three-dimensional structural parts, and has the characteristics of automation, high efficiency, good stability, high recognition accuracy, and good adaptability, meeting the flexible and intelligent operation requirements of welding robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a schematic diagram of the structure of the device of the present invention, a welding robot, and a DLP structured light camera in an assembled state; in the figure, 1, positioning device, 2, DLP structured light camera, 3, three-dimensional structure workpiece, 4, teaching pendant, 5, controller, 6, body, P1, DLP structured light camera view angle 1 position, P2, DLP structured light camera view angle 2 position, P3, DLP structured light camera view angle 3 position;

[0057] Figure 2 It is a structural schematic diagram of the device of the present invention;

[0058] Figure 3 Figures a to f are physical images of workpieces with three-dimensional structures to be welded in accordance with an embodiment of the present invention; Figures a to f are physical images of workpieces with different types of three-dimensional structures of medium and thick plates, respectively;

[0059] Figure 4 yes Figure 3 The three-dimensional structure workpiece shown is processed by the device of the present invention to obtain the corresponding point cloud model of the workpiece and the weld positioning result. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] The embodiment of the present invention discloses a device for positioning weld seams of a three-dimensional workpiece based on DLP three-dimensional vision. Figure 1 As shown, the device 1 controls a welding robot to locate welds on a three-dimensional workpiece 3 by cooperating with a DLP structured light camera 2. The welding robot includes a teach pendant 4, a controller 5, and a body 6. The welding robot is preferably a Yaskawa robot AR1440, including a teach pendant, a controller 5, a body 6, and a welding gun installed at the welding end of the body. The robot controller is preferably a Yaskawa robot controller YRC1000. The teach pendant, the controller 5, and the body 6 of the welding robot cooperate to control the robot to drive the DLP structured light camera installed at the welding end of the body to move to multiple designated positions for point cloud data collection. The collected multi-view point clouds are analyzed and processed by the workpiece point cloud model generation unit, the weld positioning unit, and the welding trajectory planning unit to obtain the spatial position and posture of each weld on the workpiece, and plan the robot's welding trajectory. The welding robot completes the welding operation of the three-dimensional workpiece according to the welding trajectory.

[0062] In the physical connection relationship, the device of the present invention is electrically connected to the DLP structured light camera and the controller respectively. The DLP structured light camera is fixed to the welding end of the robot body, and the robot body is connected to the controller and the teaching pendant in turn through electrical signals.

[0063] like Figure 2 As shown, the device of the present invention includes a workpiece point cloud model generating unit 1-1, a weld positioning unit 1-2 and a welding trajectory planning unit 1-3.

[0064] In an embodiment of the present invention, specifically, the workpiece point cloud model generation unit 1-1 is used to shoot the workpiece in a multi-perspective manner through a DLP structured light camera mounted on the welding end of the robot body, and perform multi-perspective point cloud registration in combination with the robot body joint posture information and the hand-eye calibration coefficient to generate a complete point cloud model of the workpiece.

[0065] In the embodiment of the present invention, the multi-view mode is preferably Figure 1 The three perspectives P1, P2, and P3 are shown.

[0066] The workpiece point cloud model generating unit preferably includes:

[0067] The hand-eye pose matrix acquisition module is used to calibrate the pose of the DLP structured light camera and the robot, and obtain the hand-eye pose matrix of the DLP structured light camera relative to the welding gun at the welding end of the robot body.

[0068] The robot posture reading module is used to read the posture of the welding gun at the welding end of the robot body relative to the robot base coordinate system from the robot controller in different postures of the robot

[0069] Point cloud coordinate transformation module is used to transform the workpiece point cloud P captured by the DLP structured light camera at the current viewing angle into i Transform to the robot base coordinate system middle;

[0070] The coarse registration module is used to control the robot to drive the DLP structured light camera installed at the welding end of the body to move to multiple designated positions to collect point cloud data. The robot posture reading module and point cloud coordinate transformation module are repeated for the collected multi-view point clouds to transform all the workpiece point cloud data captured from each view into the robot base coordinate system to complete the coarse registration of the workpiece model.

[0071] The precise registration module uses the improved ICP registration algorithm to accurately align the point clouds captured from multiple perspectives on the obtained coarse registration point cloud model, thereby obtaining a complete point cloud model of the workpiece. The algorithm specifically includes the following steps:

[0072] (1) In order to speed up the registration of point clouds, the voxel sampling algorithm is first used to collect point clouds from each viewpoint PC = {P1,...,P i ,...,P n} is downsampled, and the downsampled point cloud set is represented as PC′={P1′,…,P i ′,…,P n '};

[0073] (2) For each downsampled point, solve its corresponding normal and redirect the normal to point to the camera viewpoint;

[0074] (3) For the point cloud P1 collected from the first perspective, all its points are added to the fused point cloud P F For the downsampled point cloud P1′, all its points are added to the downsampled fusion point cloud P F 'middle;

[0075] (4) For the downsampled point cloud P obtained from other image perspectives i ′, first use the KD-Tree algorithm to quickly match and calculate P i Each point in P F′, for these obtained matching point pairs, set the maximum distance threshold and normal angle threshold to eliminate the wrong matching point pairs;

[0076] (5) Construct the coordinate transformation equation of the matching point pair set and use the least squares method to iteratively calculate the optimal rotation matrix R and translation matrix T;

[0077] (6) Repeat steps (4) to (5) to solve the point cloud P at each viewing angle i ′ and P F The transformation matrix T between i , then, through the transformation matrix T i , the point cloud P i ′Integrate into point cloud P F ′, and the point cloud P i Integration into point cloud P F middle;

[0078] (7) For the multi-view fusion point cloud P obtained in step 6 F , which contains a large number of repeated redundant points, which will affect the speed of subsequent processing algorithms; therefore, the multiple of the point cloud P1 resolution N is used as the distance threshold. F For each point in , if there are other neighboring points within the distance threshold, these neighboring points are regarded as redundant points and are eliminated, and finally a complete point cloud model of the workpiece can be obtained.

[0079] In an embodiment of the present invention, the weld positioning unit 1-2 is used to obtain effective positioning of multiple welds on the model-free medium-thick plate three-dimensional structural member based on the complete point cloud model.

[0080] Specifically, the weld positioning unit includes: a point cloud data preprocessing module for performing through filtering, noise filtering, and work platform point cloud removal on the point cloud data; wherein the point cloud data preprocessing module is specifically used to:

[0081] Through filtering: retain the point cloud within the limited range of each dimension in the original point cloud data, and eliminate redundant data such as background point sets;

[0082] Noise filtering: Use statistical filtering algorithms to filter point cloud data to resist interference from factors such as environmental noise, low-key system noise, and edge discrete points;

[0083] Work platform point cloud elimination: Use the Ransac algorithm to fit and eliminate the work platform point cloud where the workpiece is located, retaining only the point cloud model of the workpiece to be welded;

[0084] The feature point set recognition module is used to filter out high-curvature feature point sets for subsequent weld identification from the acquired complete point cloud model. Specifically, the feature point set recognition module is used to:

[0085] Covariance analysis: For any point p in the point cloud i , search for its support radius r support All neighboring points Q i ={q i1 ,q i2 ,…,q in}, and p i Set as the center and calculate the covariance matrix of the local point cloud:

[0086]

[0087] Among them, d ij is the neighborhood point q ij To the center point p i The Euclidean distance of

[0088] Eigenvalue solution: Use SVD operation to decompose the covariance matrix M and find the corresponding eigenvalues ​​λ1, λ2, and λ3, where λ1>λ2>λ3;

[0089] Feature point setting: Calculate the value of λ3 / λ2 and compare it with the set feature threshold. If it is greater than the threshold, it is set as a feature point;

[0090] Feature point set screening: Repeat covariance analysis and eigenvalue solution to complete the screening of workpiece feature point sets;

[0091] The feature point concavity and convexity analysis module is used to perform geometric morphology analysis on the neighborhood of each high-curvature feature point, complete concavity and convexity judgment, and filter out the feature point set with concave attributes. The feature point concavity and convexity analysis module is more specifically used to:

[0092] Directed normal calculation: For a certain high-curvature feature point, we first perform principal component analysis on the point set within its R radius neighborhood, take the eigenvector corresponding to the minimum eigenvalue as the normal of the feature point, and redirect the normal direction to point to the camera viewpoint, thereby eliminating the bidirectional ambiguity of the normal.

[0093] Feature point centroid analysis: Calculate the centroid of the neighborhood point set of the feature point, and define the vector from the feature point to the centroid as the concave-convex direction vector d of the point;

[0094] Convexity determination of feature points: Calculate the angle between the feature point's orientation normal and the concave-convex direction vector d. When the angle between normal and d is greater than 90°, the feature point has a convex attribute. Conversely, if the angle between normal and d is less than 90°, the feature point has a concave attribute.

[0095] Traversal and repeated analysis: For other high curvature point sets, the process of oriented normal calculation, feature point centroid analysis, and feature point concavity determination is repeated in sequence to complete the concavity and convexity attribute determination of all feature points;

[0096] The feature line extraction module is used to extract multiple feature lines from the acquired concave attribute feature point set using the Ransac algorithm; wherein the feature line extraction module is more specifically used to:

[0097] Recording of feature line interior points: Randomly select two feature points from the extracted concave attribute feature point set for straight line fitting, and calculate the distance from other feature points to the fitted straight line. If the distance is less than the set threshold, it is recorded as the interior point of the feature line;

[0098] Fitting line analysis: Repeat the recording of the inner points of the characteristic line, record the inner points and their number of each fitting line, and repeat the iteration The number of times ends, where z is the probability of obtaining a good sample and w is the probability of an inlier point in the point set;

[0099] Fitting and elimination: Select the result with the largest number of inliers during the iteration process to complete the fitting of this round of feature lines, and eliminate the inliers of the feature lines from the original feature point set;

[0100] Traversal fitting: the process of recording the points inside the feature line, analyzing the fitted line, and fitting and eliminating points is repeated continuously until the number of points inside the fitted feature line fails to meet the set minimum threshold, and the fitting of all feature lines is completed;

[0101] The feature line segmentation module is used to segment each feature line using the Euclidean clustering algorithm to obtain multiple point clusters with independent spatial positions. The feature line segmentation module is specifically used to:

[0102] Inclusion into clustering: For a set of points within a straight line, randomly select a seed point and use the KD-Tree algorithm to search the R radius neighborhood of the seed point. If there are other points in the neighborhood, they and the seed point are included in the cluster Q;

[0103] Traversal clustering: randomly select new seed points in the cluster Q and repeat the process of clustering until the number of points in cluster Q no longer increases, then the clustering ends; if the cluster point cloud cluster Q meets the set minimum number of points, then the clustering result is retained and the inner points of cluster Q are removed from the original point cloud;

[0104] Multi-cluster inclusion: Randomly select new seed points from the remaining point cloud and repeat the process of inclusion, traversal, and multi-cluster inclusion to divide the original point cloud into multiple point clusters with independent spatial positions.

[0105] The weld positioning module is used to cluster a certain segment point, project all points within the cluster onto the characteristic line according to the cluster, and obtain the two endpoints of the weld through a sorting algorithm. The weld positioning module is more specifically used to:

[0106] Set projection: For each cluster point cloud cluster obtained by segmentation, the internal point set of the point cloud cluster is projected onto the line according to the fitting parameters of the line to which it belongs;

[0107] Solve the current projection endpoint: Sort the projection point set on the projection line and find the starting and ending projection points of the point cloud cluster in the direction of the line. These two projection endpoints are the endpoints of the weld.

[0108] Solve all projection endpoints: Repeat the process of collective projection and current projection endpoint solution to complete the confirmation of all cluster point cloud cluster projection endpoints, and then all weld endpoints can be solved.

[0109] In the embodiment of the present invention, the welding trajectory planning units 1-3 are used to solve the welding gun posture corresponding to each weld seam and plan the welding trajectory of the robot body.

[0110] Specifically, the welding trajectory planning unit preferably includes:

[0111] The discrete point sampling module is used to complete the sampling of discrete points of a certain extracted weld line segment starting from the initial point of the line segment with a fixed step size;

[0112] The discrete coordinate system construction module is used to establish a discrete coordinate system relative to the robot's base coordinate system at each discrete point. The coordinate system consists of a normal vector n, a direction vector v, and an approach vector a. The normal vector n is obtained by PCA analysis of the discrete point's neighborhood point set. The direction vector v is the tangent direction of the current discrete point on the fitted line segment. The approach vector a is obtained by cross-producting the direction vector v and the normal vector n.

[0113] The welding posture solving module is used to establish a welding gun coordinate system at the welding gun tip position, and calculate the welding posture of the welding gun at each discrete point of the weld according to the corresponding relationship between the weld posture and the welding gun posture;

[0114] The motion trajectory acquisition module is used to perform least square fitting on the welding gun trajectory to obtain the motion trajectory of the welding gun during welding of each weld.

[0115] In the actual application process of the present invention, select Figure 3 The thick plate structural part shown in the figure is treated as the welding object. According to the above implementation process of the present invention, the point cloud model of the structural part is reconstructed and the welds used to plan the welding trajectory of the robot body are extracted, such as Figure 4As shown in the implementation results, it can be seen that the present invention does not require a CAD model of the welding workpiece, has the characteristics of automation, high efficiency, good stability, high recognition accuracy, good adaptability, etc., and meets the flexible and intelligent operation requirements of welding robots.

[0116] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A DLP 3D vision-based device for locating weld seams of three-dimensional workpieces. The device achieves weld seam positioning of three-dimensional workpieces by cooperating with a DLP structured light camera and a welding robot. The welding robot includes a teaching pendant, a controller, and a body. The device is characterized in that: The device comprises: The workpiece point cloud model generation unit is used to capture the workpiece from multiple perspectives using a DLP structured light camera mounted on the welding end of the robot body. This unit then performs multi-perspective point cloud registration based on the robot body's joint pose information and hand-eye calibration coefficients to generate a complete point cloud model of the workpiece. A weld positioning unit, configured to obtain effective positioning of multiple welds on a model-free medium-thick plate three-dimensional structural component based on the complete point cloud model; The welding trajectory planning unit is used to solve the welding gun posture corresponding to each weld seam and plan the welding trajectory of the robot body; The weld positioning unit comprises: Feature point set recognition module, used to filter out high curvature feature point sets for subsequent weld identification from the acquired complete point cloud model; The feature point concavity and convexity analysis module is used to perform geometric morphology analysis on the neighborhood of each high-curvature feature point acquired, complete concavity and convexity judgment, and screen out the feature point set with concave attributes; The feature line extraction module is used to extract multiple feature lines from the acquired concave attribute feature point set using the Ransac algorithm; The feature point concavity and convexity analysis module is specifically used for: For a certain high curvature feature point obtained, first perform principal component analysis on the point set within its R radius neighborhood, take the eigenvector corresponding to the minimum eigenvalue as the normal of the feature point, and redirect the normal direction to point to the camera viewpoint, thereby eliminating the bidirectional ambiguity of the normal; Calculate the centroid of the neighborhood point set of the feature point, and define the vector from the feature point to the centroid as the concave-convex direction vector d of the point; Calculate the angle between the feature point's orientation normal and the concave-convex direction vector d. If the angle between normal and d is greater than 90°, the feature point has a convex attribute. If the angle between normal and d is less than 90°, the feature point has a concave attribute. For other high curvature point sets, the process of oriented normal calculation, feature point centroid analysis, and feature point concavity determination is repeated in sequence to complete the concavity and convexity attribute determination of all feature points. The feature line extraction module is specifically used for: On the extracted concave attribute feature point set, two feature points are randomly selected for straight line fitting, and the distance from other feature points to the fitted straight line is calculated. If the distance is less than the set threshold, it is recorded as the inner point of the feature line; Repeat the recording of the interior points of the characteristic line in a loop, and record the interior points and their number of the fitted straight line each time; The result with the largest number of inliers in the iterative process is selected to complete the fitting of this round of feature lines, and the inliers of this feature line are removed from the original feature point set; The process of recording the inner points of the characteristic lines, analyzing the fitted lines, and fitting and eliminating them is repeated continuously until the number of inner points of the fitted characteristic lines fails to meet the set minimum threshold, and the fitting of all characteristic lines is completed.

2. The three-dimensional structure workpiece weld positioning device according to claim 1, characterized in that: The workpiece point cloud model generating unit includes: A hand-eye pose matrix acquisition module is used to calibrate the poses of the DLP structured light camera and the robot, and obtain the hand-eye pose matrix of the DLP structured light camera relative to the welding gun at the welding end of the robot body; The robot posture reading module is used to read the posture of the welding gun at the welding end of the robot body relative to the robot base coordinate system from the robot controller in different postures of the robot; Point cloud coordinate transformation module, used to transform the workpiece point cloud captured by the DLP structured light camera at the current viewing angle into the robot base coordinate system; The coarse registration module is used to control the robot to drive the DLP structured light camera installed at the welding end of the body to move to multiple designated positions to collect point cloud data. The robot pose reading module and point cloud coordinate transformation module are repeated for the collected multi-view point clouds to transform all the workpiece point cloud data captured from each view into the robot base coordinate system to complete the coarse registration of the workpiece model. The precise registration module samples the obtained coarse registration point cloud model and uses the improved ICP registration algorithm to complete the precise alignment of the point clouds shot under multiple perspectives, thereby obtaining a complete point cloud model of the workpiece.

3. The three-dimensional structure workpiece weld positioning device according to claim 1, characterized in that: The weld positioning unit further comprises: The point cloud data preprocessing module is used to perform straight-through filtering, noise filtering, and work platform point cloud removal on the point cloud data; the feature line segmentation module is used to segment each feature line using the Euclidean clustering segmentation algorithm to obtain multiple point clusters with independent spatial positions; The weld positioning module is used to cluster a certain point, project all points inside the cluster onto the characteristic line according to the characteristic line where the cluster is located, and obtain the two endpoints of the weld through a sorting algorithm.

4. The three-dimensional structure workpiece weld positioning device according to claim 3, characterized in that: The point cloud data preprocessing module is specifically used for: Through filtering: retain the point cloud within the limited range of each dimension in the original point cloud data, and eliminate redundant data such as background point sets; Noise filtering: Use statistical filtering algorithms to filter point cloud data to resist interference from factors such as environmental noise, low-key system noise, and edge discrete points; Work platform point cloud elimination: Use the Ransac algorithm to fit and eliminate the work platform point cloud where the workpiece is located, retaining only the point cloud model of the workpiece to be welded.

5. The three-dimensional structure workpiece weld positioning device according to claim 3, characterized in that: The feature point set recognition module is specifically used for: Covariance analysis: For any point p in the point cloud i , search for its support radius r support All neighboring points Q within i ={q i1 ,q i2 ,…,q in }, and p i Set as the center and calculate the covariance matrix of the local point cloud: where d ij is the neighborhood point q ij To the center point p i The Euclidean distance of Eigenvalue solution: Use SVD operation to decompose the covariance matrix M and find the corresponding eigenvalues ​​α1, α2, and λ3, where λ1>λ2>λ3; Feature point setting: Calculate the value of λ3 / λ2 and compare it with the set feature threshold. If it is greater than the threshold, it is set as a feature point; Feature point set screening: Repeat covariance analysis and eigenvalue solution to complete the screening of the workpiece feature point set.

6. The three-dimensional structure workpiece weld positioning device according to claim 3, characterized in that: The feature line segmentation module is specifically used for: Inclusion into clustering: For a set of points within a straight line, randomly select a seed point and search the neighborhood of the seed point within a radius of R. If there are other points in the neighborhood, include them and the seed point into the cluster Q; Traversal clustering: randomly select new seed points in cluster Q and repeat the process of clustering until the number of points in cluster Q no longer increases, then the clustering ends; Multi-cluster inclusion: New seed points are randomly selected from the remaining point cloud, and the process of inclusion, traversal, and multi-cluster inclusion is repeated to divide the original point cloud into multiple spatially independent clusters.

7. The three-dimensional structure workpiece weld positioning device according to claim 3, characterized in that: The weld positioning module is specifically used for: Set projection: For each cluster point cloud cluster obtained by segmentation, the internal point set of the point cloud cluster is projected onto the line according to the fitting parameters of the line to which it belongs; Solve the current projection endpoint: Sort the projection point set on the projection line and find the starting and ending projection points of the point cloud cluster in the direction of the line. These two projection endpoints are the endpoints of the weld. Solve all projection endpoints: Repeat the process of collective projection and current projection endpoint solution to complete the confirmation of all cluster point cloud cluster projection endpoints, and then all weld endpoints can be solved.

8. The three-dimensional structure workpiece weld positioning device according to claim 1, characterized in that: The welding trajectory planning unit includes: The discrete point sampling module is used to complete the sampling of discrete points of a certain extracted weld line segment starting from the initial point of the line segment with a fixed step size; Discrete coordinate system construction module, used to establish a discrete coordinate system relative to the robot base coordinate system at each discrete point. The coordinate system consists of a normal vector, a direction vector, and a proximity vector; The welding posture solving module is used to establish a welding gun coordinate system at the welding gun tip position, and calculate the welding posture of the welding gun at each discrete point of the weld according to the corresponding relationship between the weld posture and the welding gun posture; The motion trajectory acquisition module is used to perform least square fitting on the welding gun trajectory to obtain the motion trajectory of the welding gun during welding of each weld.

Images